OpenAI’s new voice AI model gpt-4o-transcribe lets you add speech to your existing text apps in seconds

Openai’s voice EU models won it Problem before the actor Scarlett JohanssonBut this does not stop the company’s suspension to develop their proposals in this category.

Today, Announced on Chatgpt Maker The three, all new ownership models called GPT-4O-TransCRIBE, GPT-4O-mini-transcription and GPT-4O-mini-ttsOriginally available for third-party software developers to set up their application on the application programming interface (API) and a special demo website Openai.fmIndividual users can get for a limited trial and entertainment.

https://www.youtube.com/watch?v=lxb0l16isac

Moreover, GPT-4O-mini-TTS model sounds, accents, pitch, tons and other vocal qualities are asked to address any concerns, including several selections to imitate any user’s voice.The company previously rejected Johansson’s conditionHowever, in any case, the ima of silence has lowered the sound selection). Now it is to decide how they want to play AI sounds when talking back to the user.

Jeff Harris, a demo, a demo, which is delivered with video survey, can be sounded like a noise or zen, a quiet yoga teacher.

Discover and clear new opportunities in the GPT-4O database

Models are options The current GPT-4o model Openai has been withdrawn in May 2024 Currently, ChatGpt for many users enhances text and sound experience, but the company has taken this base model and gave additional information taught with additional information to make transcription and speech in transcription and speech. The company was not clarified that the models could come to Chatgpt.

“ChatGpt has a little different requirements for waiting for this startup to these models that I expected to pass these models to these models,” Harris said.

This is designed to offer a two-year-old whispered whispering open-spoken text model, different accents and noisy environments in various accents and noisy environments, variable speech and noisy performance.

The company placed a schedule on the website, determining the words in 33 languages ​​of GPT-4O-4O-transcribe models, compared to whispering – with an effective effective in English.

“These models include noise cancellation and semantic voice activity detector, which helps to determine that a speaker has completed a thought and increases the accuracy of the transcription,” Harris said.

Harris Venturebeat said that the new GPT-4O-Transcribe model family was not designed to label and distinguish between “dialia” or various speakers. Instead, it is designed primarily to get one (or multiple votes like an entrance channel) and respond to all entries with a single exit voice in this interaction.

The company is conducting a competition to find the most creative samples of the general public to find the most creative samples of online sharing online sharing by using Openai.FM. @Openai account x. The winner began to buy a special teenager radio with Openai logo, which is one of the only three of the Openai product, the platform Olivier God.

An audio applications gold mine

Strengthens are especially well adapted for applications such as Customer Call Centers, Note Transcriptions and AI-Energy Assistants.

Effective, The new launched agents SDK From last week, the text-based large language models with the “nine code line” with the “nine code line” with the “nine code line”, a presentation that announces new models (installed above).

For example, an e-commerce application installed in GPT-4O, now adds these new models to the queue-based user questions like “Talk about my last orders” when adding these new models.

“For the first time, the text of the text, which allows developers to enter the sound of constantly entering the sound and the text of the text, the text of the text makes the conversations feel more natural,” Harris said.

However, for those looking for a low retardant, real-time searching for AI voice experiences, Openai recommends using speech speaking models in Realtime API.

Price and availability

The new models can be obtained immediately through Openai’s API, the price is as follows:

• GPT-4O-TransCRIBE: 1M Audio Input Tokens $ 6.00 ($ 0.006 per minute)

• GPT-4O-mini-transcription: $ 3.00 per $ 1 million audio access tokens ($ 0.003 per minute)

• GPT-4O-MINI-TTS: $ 0.60 for $ 1 million text access $ 12.00 (0.015 per minute)

However, according to AI transcription and speech, special speech, according to AI firms, they reach a violent time in competition forever Onelabs offers the new Scribe model This supports dialiatisation and is a 3.3% reduced error rate and $ 0.40 per hour (approximately equivalent) per hour (approximately equivalent).

Another start, Hume AI offers a new model Octave TTS Based on user instructions, not pre-determined sounds – pronunciation and emotional infection and even customization of word level – not prescribed sounds. The price of Octave TTS is not directly comparable, but it offers a proposal to increase the sound and expenses for 10 minutes

Meanwhile, more advanced voice and speech models include a society, including open sources, including Orfeus 3B, which exists with the Apache 2.0 license that allowsThe developers do not have to pay no cost to manage it – when proper apparatus or cloud servers.

Industrial setting and early results

Several companies, according to the phrase shared by VentureBeat and Openai, several companies that provide information on important developments in the audible AI performance.

A company focused on the automation of property management, Eliseai, Openai, found a more natural and emotional rich interior relations with the text model of speech.

More attractive, repair work is being carried out, AI-powered leasing, maintenance and tour planning, more attractive, higher tenant satisfaction and rich resolution rates.

The decagon, which builds AI-free voice experiences, using Openai’s speech recognition model I have been improving 30% in the accuracy of transcription.

This increase in accuracy has allowed the decagon to implement the AI ​​agents in real world scenarios in even noisy environments. The integration process was quickly with the decagon, which includes a new model in its system in one day.

All the reactions made to the latest issue of Openai were not hot. Dawn EU App Analytics program co-founder I Hylakam (@benhylak)An apple man’s interface designer is placed in the X when the models are seen in the X, “Real-time felt like retreat”, offers to move away from the previous focus of the low-delayed conversation via Chatgept.

In addition, the starting X (earlier Twitter) was before the beginning of early leakage. TestKatalog news (@TStingCatalog) sent Details On The New Models Several Minutes Before The Official Announcement, Listing The Names of GPT-4O-Mini-Tts, GPT-4O-Transcribe, and GPT-4O-Mini-Transcribe. The leak was transferred to @ Stiventhedev and quickly pulled the article quickly.

However, Openai plans to investigate special voice opportunities, while Openai continues to clean and security and use a responsible AI. Outside Audio, Openai is also investing in multimodal AI, including video to provide more dynamic and interactive agent-based practices.